In the global AI arena, open-source models are increasingly moving beyond challengers and into the mainstream. Following Chinese AI companies such as DeepSeek, Z.ai and Moonshot AI, U.S. companies are also putting forward open-weight models that disclose weights. That appears to increasingly threaten OpenAI and Anthropic, which have led the AI field with closed models.

Chinese AI startup Moonshot AI is unveiling Kimi K3, a large language model based on open weights that matches the performance of U.S. frontier models. K3 is drawing attention for surpassing Anthropic Opus 4.8 on major benchmarks. After China’s Kimi K3 ranked first in a coding performance evaluation, David Sacks warned that U.S. AI competitiveness is weakening.

ㆍAfter GLM-5.2, another ultra-powerful Chinese AI model is coming...this time Moonshot AI’s Kimi K3 draws attention ㆍ"It won the benchmark, but not sure"…Silicon Valley’s view of China’s Kimi K3 ㆍChina’s 'Kimi K3' ranks No. 1 in coding evaluation…David Sacks criticises U.S. AI regulation

In the United States, more AI companies are focusing on open source. Thinking Machines Lab, an AI startup led by former OpenAI executive Mira Murati (미라 무라티), introduced its first in-house AI model, Inkling, as an open-weight model that outside developers and companies can download and modify themselves.

ㆍThinking Machines bets on customised AI for companies…unveils first open-weight model 'Inkling'

Interest in open-source models is also having a significant impact on global stock markets. After Moonshot AI recently announced the Kimi K3 model, shares of semiconductor companies such as Nvidia and Micron plunged. TSMC and ASML posted strong quarterly results, but that was not enough to reverse the decline.

ㆍ[Tech Inside] The truth behind semiconductor share declines sparked by Chinese AI models ㆍSelling in U.S. semiconductor stocks spreads to Asia…SoftBank drops 8 percent

The following is a roundup of moves by companies in and outside South Korea related to AI.

OpenAI officially entered the hardware market by launching the Codex Micro keyboard that links to its AI coding assistant service Codex.

ㆍOpenAI unveils $230 keyboard linked to Codex…enters hardware market

Anthropic set up a joint venture called Ode with Blackstone, Hellman & Friedman and Goldman Sachs to handle the adoption and deployment of enterprise AI.

ㆍAnthropic strengthens enterprise AX business with joint venture Ode

Google added a feature in AI Mode that connects some external apps so users can continue tasks directly. The feature will be applied first to U.S. users, and supported apps at launch are Instacart, Canva and YouTube. Google changed the brand name of its AI-based research tool NotebookLM to Gemini Notebook and added a code-execution feature for data analysis. Google also introduced a personal AI avatar feature in its video creation tool Google Vids that reflects a user's face and voice. Shares of Google parent Alphabet fell after reports about a delay in the release of its flagship AI model, Gemini 3.5 Pro. Google is speeding up its push into South Korea’s AI market by highlighting Gemini Enterprise’s full-stack architecture and compatibility with third-party models.

ㆍGoogle adds app connectivity to AI Mode…supports Instacart, Canva and YouTube ㆍNotebookLM reborn as Gemini Notebook...adds code execution feature for data analysis ㆍGoogle Vids supports creating a personal AI avatar with selfies and voice ㆍAlphabet shares fall on reports of a delay in launching 'Gemini 3.5 Pro' ㆍGoogle: "A head-on challenge in South Korea’s AI market with Gemini Enterprise"

Apple received approval to launch in China by linking Alibaba’s Qwen AI to its AI platform Apple Intelligence.

ㆍApple Intelligence enters China by linking Alibaba Qwen AI

Nvidia launched a small open-source model for 'physical AI' that runs in the real world, such as robots. Cosmos 3 Edge has 4 billion parameters and can run on customers’ own computers rather than in data centres.

ㆍNvidia launches small physical AI model 'Cosmos 3 Edge'

Alibaba Group Holding’s chip design unit T-Head is open-sourcing its own software stack. The move aims to challenge Nvidia’s dominance of the CUDA ecosystem.

ㆍAlibaba T-Head open-sources its own AI chip software

Intel is rolling out Google Cloud’s Gemini Enterprise across the company and moving to automate semiconductor design and internal work. Intel plans to use it to accelerate workflows across operations, engineering, supply chains and marketing, and to shorten chip design cycles.

ㆍIntel adopts Gemini Enterprise companywide…pushes semiconductor design automation

Cadence Design Systems unveiled Orastack, an agentic AI system that supports printed circuit board (PCB) design and testing for electrical engineers.

ㆍCadence unveils agentic AI 'Orastack'...supports PCB design

SpaceX is in talks with the U.S. Defense Department to provide data centre capacity for running AI models. SpaceX has in recent months been accelerating its expansion into cloud computing, including signing similar contracts with Anthropic and Google. Anthropic is in early-stage talks on renting computing power from Meta.

ㆍSpaceX in talks with the Defense Department to provide AI data centres worth billions of dollars ㆍAnthropic begins discussions on renting Meta computing power

Bespin Global, a company specialising in AI services and solutions, built a core function called AICSR (AI CSR Assistant) for a smart AI messenger project aimed at innovating next-generation customer consultation at AIA Life. Oracle added a new AI-native builder environment to its AI development tool, AI Agent Studio, to allow users to develop and operate AI agentic apps within its enterprise integrated business system, Fusion Applications.

ㆍBespin Global builds AIA Life generative AI consultation assistant on private cloud ㆍOracle adds agentic app development environment to AI Agent Studio

Chinese President Xi Jinping proposed expanding AI cooperation with developing countries and urged against over-interpreting AI security.

ㆍXi proposes China AI cooperation for developing countries…warns against excessive security focus

The government will roll out a free and unlimited AI service for all citizens, called Everyone’s AI, within this year. The government says Everyone’s AI is needed to reduce reliance on foreign AI. Some point out that with private-sector free AI already widely used, Everyone’s AI needs a differentiated role, especially as it will involve deploying public GPUs.

ㆍWith many free AI services already…will the government’s 'Everyone’s AI' work?